Algorithm Fusion in Novelty Detection - Inria - Institut national de recherche en sciences et technologies du numérique Access content directly
Conference Papers Year : 2014

Algorithm Fusion in Novelty Detection


Algorithm fusion has received significant attention in the machine learning community in supervised learning mode but it appears little has been done at this point in a novelty detection framework. This paper examines the merit of a fusion strategy wherein metrics from multiple algorithms are treated as entries of a vector whose probability density is subsequently estimated and used for detection. In the present paper the framework is investigated using two algorithms: 1) a robust version of a whiteness test on Kalman filter innovations and 2) a robust version of a scheme that operates with residuals obtained from an orthogonality test. The density estimation part of the process is replaced by the Kernel PCA algorithm which provides a decision boundary without having explicit density estimates. The fused scheme is implemented in a change detection format and is show to provide notable improvements over the use of either algorithm independently.
Fichier principal
Vignette du fichier
0259.pdf (128.33 Ko) Télécharger le fichier
Origin : Files produced by the author(s)

Dates and versions

hal-01021206 , version 1 (09-07-2014)


  • HAL Id : hal-01021206 , version 1


Dionisio Bernal. Algorithm Fusion in Novelty Detection. EWSHM - 7th European Workshop on Structural Health Monitoring, IFFSTTAR, Inria, Université de Nantes, Jul 2014, Nantes, France. ⟨hal-01021206⟩
85 View
71 Download


Gmail Facebook Twitter LinkedIn More